TalkChess.com

[Eelco de Groot]

This is from the very first testgame, played against Stockfish. I can see there are still a lot of issues, but I don't think that starting tuning only after implementation of several features, by definition would be wrong. It is part of a more simulated annealing approach. In other words, a lot of work But the point of the testposition is finding some keypoints to start the tuning and here clearly Rainbow Serpent is underestimating the two passed pawns of Black, combined with its bishop pair. In the game it only saw itself down 0.12 pawns at move 21, where I think the game is already decided, in a higher sense. Not easy for the computer to see that but that is the test, not a best move or move to avoid, but more "how big is the score for black and does it go up in time?" Stockfish does have a large eval already even at the blitz timecontrol I used. I think there were several eval features combining to give a wrong evaluation here, or also some wrong continuations possible. Several things changed after this game. For instance the rank of the passed pawn is now calculated from Rank_1, not Rank_2. This means passed pawns still on the second rank now get a more realistic bonus where in Stockfish for instance it is almost nothing. A further slight increase in the base calculation even if the passed pawn is blocked. A bonus for having the bishop pair with every passed pawn. Isolated pawns only get a very small penalty for being isolated if they are passed, and if they are passed no more penalty for being isolated only on one file to the left or right. Further: changes to the eval of doubled isolated pawns. This just to show to the innocent Ivanhoes how important it is to have a good eval. What do they make of this position with a random evaluation. Not much I dare predict.

This the eval of Build 053 after most of the changes above but not yet any of the changes in pawn evaluation that are in pawn.cpp, only the changes to passed pawns that are in the dynamic passed pawn eval in evaluate.cpp. Just one thread used but it is clear how much these changes introduced already influence the eval of this position. Not a clear win but much better than +0.12.


[D]r7/1p2bp1p/p1k1bp2/4p3/8/P1N1R3/1PP3RP/4K3 b - -

Engine: Rainbow Serpent 2.2.4 one thread build 053 (256 MB)
by Tord Romstad, Marco Costalba and Joona

11/11 0:01 +1.49++ 21...Bc5 22.Re4 f5 23.Rh4 Rh8 24.Ke2 h6
25.Rh5 Bd4 26.Rh4 Bxc3 (635.431) 391

11/12 0:01 +1.02-- 21...Bc5 22.Re4 f5 23.Rxe5 Bd4
24.Ree2 Bh8 25.h3 Bxc3+ 26.bxc3 Bc4
27.Re5 Bf1 (682.681) 367

11/13 0:01 +1.17 21...Bc5 22.Re4 Bd4 23.Rd2 f5
24.Rexd4 exd4 25.Ne2 Rg8 26.Nxd4+ Kc5
27.Kf2 Bd5 28.Nxf5 Rg4 29.Ne7 Rc4 (705.632) 358

12/15 0:05 +1.25++ 21...Bc5 22.Re4 h6 23.Rd2 Rg8 24.Rh4 Rg1+
25.Ke2 f5 26.Rd8 Rg2+ (2.014.968) 363

12/15 0:06 +1.32 21...Bc5 22.Re4 h6 23.Rd2 Rg8 24.Ne2 Bd5
25.Rh4 f5 26.Rxh6+ Be6 27.Rd3 Rf8 (2.188.628) 360

13/15 0:06 +1.12 21...Bc5 22.Ree2 Bd4 23.Rg7 Rh8
24.Ne4 Bxb2 25.c3 Bxa3 26.Nxf6 Be7
27.Nxh7 Bh4+ 28.Kd2 Rd8+ 29.Ke3 Bf5
30.Rxf7 Bg6 31.Re7 (2.503.623) 361

14/15 0:07 +1.20++ 21...Bc5 22.Ree2 Bd4 23.Rg7 Rh8
24.Ne4 f5 25.c3 fxe4 26.cxd4 f5
27.dxe5 Kd5 28.Rd2+ Kxe5 29.Rxb7 f4
30.Ra7 Bd5 31.a4 (2.633.078) 364

14/15 0:07 +1.28++ 21...Bc5 22.Ree2 Bd4 23.Rg7 Rh8
24.Ne4 Bxb2 25.c3 Bxa3 26.Nxf6 Bd6
27.Ne4 f5 28.Nf6 (2.729.139) 371

14/15 0:07 +1.39 21...Bc5 22.Re4 h6 23.Rh4 Rh8 24.Re4 Bd6
25.Kf2 f5 26.Re1 f4 27.Rg7 Be7
28.Kf3 Bd6 29.Kf2 (3.029.662) 380

15/15 0:08 +1.16-- 21...Bc5 22.Re4 h6 23.Rg3 Bf2+
24.Kxf2 f5 (3.272.550) 382

15/18 0:10 +1.45 21...Bc5 22.Re4 Bd4 23.Rh4 Rh8
24.Rd2 Bxc3 25.bxc3 f5 26.Rh6 Kc5
27.Re2 Kd5 28.Rd2+ Kc4 29.Re2 e4
30.Kd2 Kd5 31.Rf2 e3+ 32.Kxe3 Ke5 (4.261.343) 409

16/18 0:10 +1.45 21...Bc5 22.Re4 Bd4 23.Rh4 Rh8
24.Rd2 Bxc3 25.bxc3 f5 26.Rh6 Kc5
27.Re2 Kd5 28.Rd2+ Kc4 29.Re2 e4
30.Kd2 Kd5 31.Rf2 e3+ 32.Kxe3 Ke5 (4.365.405) 413

17/23 0:41 +1.52 21...Bc5 22.Re4 Rh8 23.Kd2 Bd4
24.Rxd4 exd4 25.Ne2 Kd5 26.Nf4+ Kd6
27.Re2 Re8 28.Re4 h6 29.Rxd4+ (15.871.013) 383

18/25 0:49 +1.26-- 21...Bc5 22.Re4 Rh8 23.Rd2 Rg8
24.Ne2 Bd5 25.Rh4 Rg2 26.Rxh7 f5
27.Rh8 Be4 28.h4 f4 29.Rc8+ Kb6
30.h5 Bxa3 31.bxa3 (19.117.912) 389

18/29 1:16 +1.53 21...Bc5 22.Re4 Rh8 23.Kd2 Bd4
24.Rxd4 exd4 25.Ne2 Kd5 26.Rf2 Ke5
27.Nf4 Bf5 28.Nd3+ Bxd3 29.Kxd3 f5
30.Re2+ Kf6 31.Kxd4 Rd8+ 32.Ke3 f4+
33.Kxf4 Rd4+ 34.Ke3 Rh4 (35.022.524) 457

19/29 1:16 +1.53 21...Bc5 22.Re4 Rh8 23.Kd2 Bd4
24.Rxd4 exd4 25.Ne2 Kd5 26.Rf2 Ke5
27.Nf4 Bf5 28.Nd3+ Bxd3 29.Kxd3 f5
30.Re2+ Kf6 31.Kxd4 Rd8+ 32.Ke3 f4+
33.Kxf4 Rd4+ 34.Ke3 Rh4 (35.257.288) 458

20/31 1:22 +1.26 21...Bc5 22.Re4 Rh8 23.Kd2 Bd4
24.Rxd4 exd4 25.Ne2 Kd5 26.Rf2 Ke5
27.Nf4 f5 28.Nd3+ Kd5 29.Nb4+ Kc5
30.Nd3+ Kc4 31.b3+ Kd5 32.Nb4+ Kd6
33.Rf4 a5 34.Nd3 Kd5 (39.440.475) 477

21/44 5:17 +1.34++ 21...Bc5 22.Re4 Rh8 23.Rd2 b5 24.Rd3 Bd6
25.a4 Bf5 26.Rde3 Bxe4 27.Nxe4 f5
28.Nxd6 Kxd6 29.axb5 axb5 30.Rc3 Ra8 (203.196.609) 640

21/44 6:14 +1.42++ 21...Bc5 22.Re4 Rh8 23.Rd2 b5 24.Rd3 Bd6
25.a4 Bf5 26.Rde3 Bxe4 27.Nxe4 f5
28.Nxd6 Kxd6 29.Rd3+ Ke6 30.axb5 axb5
31.Rd1 f4 32.c3 (243.697.573) 651

21/44 11:51 +1.54++ 21...Bc5 22.Re4 Rh8 23.a4 Bd6
24.Ree2 f5 25.Rd2 f6 26.h4 f4 27.Rh2 (413.852.931) 581

21/44 21:07 +1.57 21...Bc5 22.Rd3 f5 23.Rg5 Rh8 24.Ne2 h6
25.Rh5 a5 26.Rc3 Kb6 27.Rf3 f6
28.Rfh3 Kb5 (790.955.622) 624

22/46 28:20 +1.57 21...Bc5 22.Ree2 Bd4 23.Nd1 Bc4
24.Rd2 f5 25.b3 Bd5 26.Rg5 Be6
27.Ke2 (1.075.426.980) 632

23/49 84:41 +1.64 21...Bc5 22.Re4 Rh8 23.Rg3 Bd6
24.Rge3 (2.470.650.876) 486

24/56 94:32 +1.63 21...Bc5 22.Re4 Rh8 23.Rg3 Bd6
24.Rge3 h5 25.Re2 Bc5 26.Ra4 Bd4
27.Rd2 Bxc3 28.bxc3 f5 29.Kf2 f6
30.Rb4 f4 31.Re2 h4 32.Re1 (2.682.989.976) 472

25/70 184:20 +1.64 21...Bc5 22.Rd3 f5 23.Rh3 Rh8 24.Nd1 h6
25.Ne3 (5.635.414.902) 509

26/70 214:46 +1.65 21...Bc5 22.Re4 Rh8 23.Ke2 Bd4
24.Rh4 f5 25.Rh6 e4 26.Rg5 Bxc3
27.bxc3 Kd5 28.Rhh5 Rc8 29.Kd2 Kd6
30.Rxh7 Ke5 31.a4 (6.650.044.963) 516

27/70 237:09 +1.61 21...Bc5 22.Re4 Rh8 23.Rd2 b5 24.b4 Ba7
25.Rd3 Bb8 26.Ne2 f5 27.Rh4 f4
28.Rh6 Rg8 29.h4 e4 30.Rd1 Kb7
31.Rd4 f3 32.Rxe4 fxe2 33.Rxh7 Bg3+
34.Kxe2 (7.528.051.659) 529

28/70 515:15 +1.65 21...Bc5 22.Re4 Rh8 23.Rd2 b5 24.b4 Ba7
25.Rd3 h5 26.Rf3 Rh6 27.Ne2 Bg4
28.Rd3 Bb8 29.Ree3 Be6 30.Rd8 Bc7
31.Ra8 f5 32.Rg3 (17.399.979.241) 562

best move: Be7-c5 time: 801:00.413 min n/s: 562.817 nodes: 17.399.979.241

Also on one thread a run for comparison by Shredder 11, this is on my Vista machine and I have not a lot of engines installed but Shredder is not too bad with engames to begin with, and a full set of Shredderbases (bitbases) 3,4,5 is installed but not the Nalimov tablebases. The Shredderbases are accessed as you can see in the output from Shredder.

r7/1p2bp1p/p1k1bp2/4p3/8/P1N1R3/1PP3RP/4K3 b - -

Engine: Shredder 11 UCI (256 MB)
by Stefan Meyer-Kahlen

11/26 0:01 +0.95 21...Bc5 22.Rd3 f5 23.b4 Ba7 24.Rg7 Rh8
25.a4 h5 26.a5 Bc4 27.Rf3 Bd4 28.h4 (700.883) 590

12/30 0:02 +1.04 21...Bc5 22.Rd3 f5 23.Rg7 Rh8 24.b4 Bd4
25.Ne2 Bb2 26.a4 Bc4 27.Rd1 b5 28.a5 Bd5 (1.897.973) 636

13/29 0:04 +1.06 21...Bc5 22.Rd3 f5 23.b4 Bd4 24.Ne2 Bb2
25.a4 Bd5 26.Rg7 e4 27.Nd4+ Kc7
28.Rd2 Bc3 29.Nxf5 Bxg7 30.Rxd5 (3.263.727) 718

14/31 0:09 +1.11 21...Bc5 22.Rd3 f5 23.b4 Bd4 24.a4 f4
25.Ne2 Be3 26.Rg7 e4 27.Rc3+ Kd6
28.Rxh7 Re8 29.a5 Rd8 30.Ng3 Ke5 (7.481.719) 751 SB:1

15/36 0:33 +1.08 21...Bc5 22.Rd3 f5 23.Rg5 Rh8 24.Rh5 h6
25.Ne2 a5 26.Ng3 f4 27.Ne4 Bd4
28.Rxd4 exd4 29.Rxa5 Kb6 30.Ra4 Rd8 (26.242.490) 774 SB:4

16/37 0:56 +1.00 21...Bc5 22.Rd3 f5 23.Rg5 Rh8 24.Rh5 h6
25.Ne2 a5 26.Ng3 f4 27.Ne4 Bd4
28.Rxd4 exd4 29.Rxa5 Kb6 30.Ra4 Rd8
31.Nf6 Bf5 (44.605.053) 790 SB:18

17/39 2:24 +1.13 21...Bc5 22.Re4 Bd4 23.Rh4 Rh8
24.Rh6 f5 25.Kd2 e4 26.Rg5 Kd6 27.b4 Ke5
28.Ne2 Bb2 29.a4 f6 30.Rgh5 Rd8+
31.Ke3 (123.959.856) 857 SB:44

18/42 6:57 +1.11 21...Bc5 22.Ree2 Bd4 23.Rg7 f5
24.Nd1 Bc4 25.Reg2 Rh8 26.Rg8 Rxg8
27.Rxg8 Be6 28.Rh8 f4 29.Rxh7 e4
30.Rh4 Be5 31.Nc3 Bf5 32.Rh6+ (334.864.180) 802 SB:165

19/38 7:39 +1.15 21...Bc5 22.Ree2 Bd4 23.Rg7 f5
24.Nd1 Bc4 25.Reg2 Rh8 26.Rg8 Rxg8
27.Rxg8 Be6 28.Rh8 f4 29.Rxh7 e4
30.c3 Bb6 31.Nf2 Kd5 32.b4 Bd8 (369.072.100) 802 SB:211

20/42 11:44 +1.17 21...Bc5 22.Ree2 Bd4 23.Rg7 f5
24.Nd1 Bc4 25.Reg2 Rh8 26.Rg8 Rxg8
27.Rxg8 Be6 28.Rh8 f4 29.Rxh7 e4
30.c3 Bc5 31.b4 Bd6 (596.445.333) 846 SB:649

21/44 17:29 +1.17 21...Bc5 22.Ree2 Bd4 23.Rg7 f5
24.Nd1 Bc4 25.Reg2 Rh8 26.Rg8 Rxg8
27.Rxg8 Be6 28.Rh8 f4 29.Rxh7 e4
30.c3 Bc5 31.b4 Bd6 (862.695.495) 822
SB:1.003

22/46 25:34 +1.13 21...Bc5 22.Ree2 Bd4 23.Rg7 f5
24.Nd1 Bc4 25.Reg2 Rh8 26.c3 Bc5
27.R7g5 Be6 28.Rh5 e4 29.b4 Be7
30.Re2 Bd6 31.Kf2 (1.322.675.191) 861
SB:1.773

23/49 58:48 +1.17 21...Bc5 22.Ree2 Bd4 23.Rg7 f5
24.Nd1 Bc4 25.Reg2 Rh8 26.c3 Bc5
27.R7g5 Be6 28.Rh5 b5 29.Nf2 Bd6
30.Ke2 h6 31.Nd3 a5 (3.045.765.335) 863
SB:7.020

24/53 213:41 +1.13 21...Bc5 22.Rd3 f5 23.Rg5 Bd4 24.b4 a5
25.Kd2 Rh8 26.Ke1 Bb6 27.Rh5 Bd8
28.Ne2 axb4 29.axb4 Be7 30.c3 h6
31.Re3 Bf6 32.c4 (11.103.715.428) 866
SB:75.598

best move: Be7-c5 time: 351:13.192 min n/s: 865.720 CPU 83.7% n/s(1CPU): 1.034.313 nodes: 18.243.434.230 SB: 163.616

I am sorry guys the PV tip sharpening code seems not quite ready for inclusion in all the programs competing in Leiden But mainly because the eval seems still weak!

Eelco

[Event "1 Minutes/Game + 1 Seconds/Move"]
[Site "Engine Match"]
[Date "2012.05.06"]
[Round "1"]
[White "Rainbow Serpent 2.2.4 Build X"]
[Black "Stockfish 2.2.2 JA"]
[Result "0-1"]

1. e4 {book 0s} c5 {book 0s} 2. Nf3 {book 0s} d6 {book 0s}
3. d4 {book 0s} cxd4 {book 0s} 4. Nxd4 {book 0s} Nf6 {book
0s} 5. f3 {book 0s} e5 {book 0s} 6. Bb5+ {book 0s} Nbd7
{+0.04/19 2s} 7. Nf5 {+0.33/12 2s} a6 {+0.16/19 0s (d5)}
8. Bxd7+ {+0.04/13 2s} Qxd7 {+0.32/19 0s} 9. Bg5 {+0.36/16
3s (Ne3)} d5 {+0.36/18 3s (Qc6)} 10. Bxf6 {+0.72/15 3s}
gxf6 {+0.44/19 0s} 11. Nc3 {+0.51/15 2s (Ne3)} dxe4
{+0.40/16 3s} 12. Ne3 {+0.36/16 1s (fxe4)} Qxd1+ {+0.76/13
2s} 13. Rxd1 {+0.52/16 0s} exf3 {+1.09/17 4s (Be6)}
14. Ned5 {+0.81/14 2s} fxg2 {+1.25/17 0s (Bc5)} 15. Rg1
{+0.55/15 3s} Kd7 {+1.21/21 0s (Bc5)} 16. Nb6+ {+0.32/14
3s} Kc6 {+1.49/21 0s} 17. Nxa8 {+0.35/16 3s} Bb4 {+1.45/17
0s} 18. Rxg2 {+0.13/16 5s} Be6 {+1.57/17 1s} 19. Rd3
{+0.13/15 2s (a3)} Rxa8 {+1.49/17 4s} 20. a3 {+0.20/15 3s}
Be7 {+1.57/18 0s (Bc5)} 21. Re3 {-0.12/14 6s (Rg7)} Bc5
{+1.93/16 3s} 22. Ree2 {-0.13/15 4s (Re4)} Bd4 {+1.77/16
5s} 23. Nd1 {-0.14/16 0s} Bc4 {+1.93/17 9s} 24. Rd2
{-0.46/16 0s} f5 {+2.10/17 2s} 25. Rg7 {0.00/15 0s (b4)} h6
{+2.14/18 6s (Rh8)} 26. c3 {-0.47/12 2s (Rdg2)} Bb6
{+2.30/16 4s (Ba7)} 27. Rdg2 {-0.92/13 2s} Bd8 {+2.54/18 0s
(Rd8)} 28. Ne3 {-0.90/15 3s (Rg8)} Bh4+ {+2.78/15 4s}
29. Kd1 {-1.14/16 3s} Be6 {+3.11/16 0s} 30. Rg8 {-1.44/12
1s} Rxg8 {+3.39/16 0s} 31. Rxg8 {-1.61/12 0s} f4 {+3.75/17
2s} 32. Ng2 {-1.46/15 0s} Bg5 {+3.83/19 3s} 33. h4
{-1.27/17 0s} Bf6 {+3.91/19 3s (Be7)} 34. Ke2 {-1.17/16 7s}
Kd5 {+4.12/20 0s (Bd5)} 35. Kf3 {-2.62/16 6s} h5 {+4.28/22
0s} 36. Nxf4+ {-2.62/15 4s} exf4 {+4.40/22 0s} 37. Kxf4
{-2.94/15 1s} Bxh4 {+4.44/19 0s} 38. Rh8 {-3.07/16 2s
(Rb8)} Bg4 {+4.32/18 2s (f6)} 39. Rh7 {-3.02/15 5s (b3)}
Ke6 {+4.60/18 2s} 40. Rh8 {-3.35/16 1s (c4)} f6 {+4.84/15
3s (Be7)} 41. Rb8 {-2.94/14 4s (Re8+)} Bg5+ {+5.81/17 3s}
42. Kg3 {-3.46/16 0s} f5 {+6.22/17 1s} 43. Rxb7 {-3.81/16
0s (Rg8)} f4+ {+6.78/12 0s (Be3)} 44. Kh2 {-8.64/14 1s
(Kf2)} f3 {+9.85/15 1s} 45. Rb6+ {-9.78/15 0s} Kf5
{+10.90/19 0s} 46. Rb7 {-10.26/16 1s} Be3 {+14.94/18 1s}
47. Rd7 {-10.85/17 0s (Rf7+)} f2 {+17.89/17 2s} 48. Rf7+
{-13.61/18 0s} Ke6 {+21.09/20 1s} 49. Rxf2 {-16.90/17 2s
(Rf8)} Bxf2 {+20.08/12 0s} 50. Kg2 {-21.95/16 3s (b4)} Be3
{+M18/19 3s} 51. a4 {-23.17/16 0s (c4)} h4 {+M9/22 2s}
52. b4 {-22.40/13 0s} h3+ {+M8/24 1s} 53. Kg3 {-M7/25 0s}
Kf5 {+M7/26 1s} 54. b5 {-M6/38 0s} Bf4+ {+M6/32 1s} 55. Kf2
{-M5/42 0s} h2 {+M5/62 1s} 56. Kg2 {-M4/91 0s} Ke4 {+M4/100
0s (Bh3+)} 57. c4 {-M3/200 0s (a5)} Ke3 {+M3/100 0s (Bh3+)}
58. bxa6 {-M2/200 0s} Bf3+ {+M2/100 0s} 59. Kh3 {-M1/200 0s
(Kf1)} h1=Q# {+M1/100 0s} 0-1
_________________
Debugging is twice as hard as writing the code in the first
place. Therefore, if you write the code as cleverly as possible, you
are, by definition, not smart enough to debug it.
-- Brian W. Kernighan